diff --git a/README.md b/README.md index da33c6a..652251f 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,19 @@ # ComfyUI-RMBG -A sophisticated ComfyUI custom node engineered for advanced image background removal and precise segmentation of objects, faces, clothing, and fashion elements. This tool leverages a diverse array of models, including RMBG-2.0, INSPYRENET, BEN, BEN2, BiRefNet, Lucida, SDMatte, SAM, SAM2, SAM3, Florence-2, YOLOv8, Big-Lama, and GroundingDINO, while also incorporating real-time background replacement, instant VRAM unloading, and enhanced edge detection for improved accuracy. +A sophisticated ComfyUI custom node engineered for advanced image background removal and precise segmentation of objects, faces, clothing, and fashion elements. This tool leverages a diverse array of models, including RMBG-2.0, VRMBG-3.0, FeyNoBg, Finegrain Box Segmenter, INSPYRENET, BEN, BEN2, BiRefNet, Lucida, SDMatte, SAM, SAM2, SAM3, Florence-2, YOLOv8/YOLO11, Big-Lama, and GroundingDINO, while also incorporating real-time background replacement, instant VRAM unloading, and enhanced edge detection for improved accuracy. ## News & Updates +- **2026/10/20**: Update ComfyUI-RMBG to **v3.3.0** ( [update.md](https://github.com/1038lab/ComfyUI-RMBG/blob/main/update.md#v330-20261020) ) + ![V3 3 0_nodes](https://github.com/user-attachments/assets/97138882-e084-43cf-a6d2-8745981109f3) + - Added Canvas (RMBG) interactive multi-layer compositing studio with aspect-locked transform gizmos and dynamic layer slots. + - Added VRMBG (Video Background Removal) node for flicker-free sequential video matting. + - Added FeyNoBg (Ultra-Fine Matting) node with deepened Swin-L architecture for fine hairs and semi-transparent subjects. + - Added Finegrain Box Segmenter node for 1024px box-promptable matting compatible with Florence-2 and YOLO bounding boxes. + - Overhauled YOLO Segmentation with YOLO26-seg & YOLO11-seg, native RGBA cutout, and collapsible advanced options drawer. + - Rebuilt Florence-2 on a native local PyTorch architecture (100% offline, zero remote code vulnerabilities, removed matplotlib). + - Native FP16 acceleration for RMBG-2.0 and BiRefNet with automatic intermediate VRAM clearing. - **2026/09/30**: Update ComfyUI-RMBG to **v3.2.0** ( [update.md](https://github.com/1038lab/ComfyUI-RMBG/blob/main/update.md#v320-20260930) ) - ![v2.9.4_sam3](https://github.com/user-attachments/assets/9b165789-a291-4df0-a147-67cca8011262) + ![v3 2 0_sam3](https://github.com/user-attachments/assets/9b165789-a291-4df0-a147-67cca8011262) - Upgraded to new `SAM3 Multiplex` segmentation node with native `safetensors` support (`sam3.safetensors`, `sam3.1_multiplex_fp16.safetensors`) - Added `unload_model` toggle across all nodes for instant GPU VRAM release - Added New `Resolution Selector` & `Image Compare` nodes @@ -243,8 +252,11 @@ install requirment.txt in the ComfyUI-RMBG folder ``` ### 4. Manually download the models: -- The model will be automatically downloaded to `ComfyUI/models/RMBG/` when first time using the custom node. +- Most models can be automatically downloaded to `ComfyUI/models/RMBG/` on first use. VRMBG-3.0 requires manual download due to official BRIA license terms. - Manually download the RMBG-2.0 model by visiting this [link](https://huggingface.co/1038lab/RMBG-2.0), then download the files and place them in the `/ComfyUI/models/RMBG/RMBG-2.0` folder. +- Manually download the VRMBG-3.0 model by visiting official [briaai/VRMBG-3.0](https://huggingface.co/briaai/VRMBG-3.0), then download the files (`config.json`, `model.safetensors`, `model.py`, `vrmbg3_config.py`) and place them in the `/ComfyUI/models/RMBG/VRMBG-3.0` folder. +- Manually download the FeyNoBg model by visiting this [link](https://huggingface.co/1038lab/FeyNobg), then download the files (`config.json`, `FeyNobg.safetensors`, `preprocessor_config.json`) and place them in the `/ComfyUI/models/RMBG/FeyNoBg` folder. +- Manually download the Finegrain Box Segmenter model by visiting this [link](https://huggingface.co/1038lab/finegrain-box-segmenter), then download the files (`finegrain-box-segmenter.safetensors`) and place them in the `/ComfyUI/models/RMBG/finegrain-box-segmenter` folder. - Manually download the INSPYRENET models by visiting the [link](https://huggingface.co/1038lab/inspyrenet), then download the files and place them in the `/ComfyUI/models/RMBG/INSPYRENET` folder. - Manually download the BEN model by visiting the [link](https://huggingface.co/1038lab/BEN), then download the files and place them in the `/ComfyUI/models/RMBG/BEN` folder. - Manually download the BEN2 model by visiting the [link](https://huggingface.co/1038lab/BEN2), then download the files and place them in the `/ComfyUI/models/RMBG/BEN2` folder. @@ -338,6 +350,25 @@ BEN2 is a more advanced version of BEN, offering: - Support for more image types - Suitable for batch processing +## VRMBG-3.0 +VRMBG-3.0 is a dedicated temporal video and image sequence background matting model by BRIA AI: +- Temporal consistency: Autoregressive state tracking conditioned on prior frame foreground features to eliminate boundary jitter. +- Chunked video processing: Supports batch-by-batch execution on long sequences without memory fragmentation. +- Native 1024x1024 resolution inference. +- Manual model placement: Requires manual download of files from [briaai/VRMBG-3.0](https://huggingface.co/briaai/VRMBG-3.0) into `ComfyUI/models/RMBG/VRMBG-3.0/`. + +## FeyNoBg +FeyNoBg is an ultra-fine dichotomous matting model featuring a deepened Swin-L backbone (263M parameters): +- Boundary reconstruction: Excels at complex dichotomous image segmentation. +- Wispy & transparent subjects: Exceptional accuracy on wispy hair strands, animal fur, transparent glassware, lace, and motion-blurred edges with zero trimap required. +- Sub-pixel foreground edge refinement: Built-in boundary optimizer ensuring crisp, halo-free results. + +## Finegrain Box Segmenter +Finegrain Box Segmenter provides 1024px high-definition prompt-guided object matting: +- Swin-B MVANet architecture: High-resolution salient object matting optimized for 1024px images. +- Florence-2 & YOLO integration: Seamlessly connects to detection `BBOX` outputs for automated, targeted segmentation. +- Robust boundary definition: Preserves fine details, hard edges, and object textures. + ## BIREFNET MODELS BIREFNET is a powerful model for image segmentation, offering: - BiRefNet-general purpose model (balanced performance) @@ -437,6 +468,9 @@ GroundingDINO is a model for text-prompted object detection and segmentation, of ## Credits - RMBG-2.0: https://huggingface.co/briaai/RMBG-2.0 +- VRMBG-3.0: https://huggingface.co/briaai/VRMBG-3.0 +- FeyNoBg: https://huggingface.co/1038lab/FeyNobg +- Finegrain Box Segmenter: https://huggingface.co/1038lab/finegrain-box-segmenter - INSPYRENET: https://github.com/plemeri/InSPyReNet - BEN: https://huggingface.co/PramaLLC/BEN - BEN2: https://huggingface.co/PramaLLC/BEN2 @@ -464,4 +498,4 @@ GroundingDINO is a model for text-prompted object detection and segmentation, of If this custom node helps you or you like my work, please give me ⭐ on this repo! It's a great encouragement for my efforts! ## License -GPL-3.0 License +GPL-3.0 License2026-10-20-00-12-39